def __init__(self, config: Dict) -> None:
super().__init__(config)
self.name = "config_default"
self.train = BaseTrainConfig(config.pop(ConfigNamesConst.TRAIN))
self.val = BaseValConfig(config.pop(ConfigNamesConst.VAL))
self.dataset_train = BaseDatasetConfig(
config.pop(ConfigNamesConst.DATASET_TRAIN))
self.dataset_val = BaseDatasetConfig(
config.pop(ConfigNamesConst.DATASET_VAL))
self.logging = BaseLoggingConfig(config.pop(ConfigNamesConst.LOGGING))
self.saving = BaseSavingConfig(config.pop(ConfigNamesConst.SAVING))
self.optimizer = optimization.OptimizerConfig(
config.pop(ConfigNamesConst.OPTIMIZER))
self.lr_scheduler = lr_scheduler.SchedulerConfig(
config.pop(ConfigNamesConst.LR_SCHEDULER))
def __init__(self, config: Dict) -> None:
super().__init__(config)
self.name = "config_default"
self.train = trainer_configs.BaseTrainConfig(config.pop(Conf.TRAIN))
self.val = trainer_configs.BaseValConfig(config.pop(Conf.VAL))
self.dataset_train = data.BaseDatasetConfig(
config.pop(Conf.DATASET_TRAIN))
self.dataset_val = data.BaseDatasetConfig(config.pop(Conf.DATASET_VAL))
self.logging = utils.BaseLoggingConfig(config.pop(Conf.LOGGING))
self.saving = trainer_configs.BaseSavingConfig(config.pop(Conf.SAVING))
self.optimizer = optimization.OptimizerConfig(
config.pop(Conf.OPTIMIZER))
self.lr_scheduler = lr_scheduler.SchedulerConfig(
config.pop(Conf.LR_SCHEDULER))
self.mlp = MLPNetConfig(config.pop("mlp"))
def __init__(self,
config: Dict[str, Any],
*,
is_train: bool = True) -> None:
super().__init__(config)
self.name = "config_ret"
self.dim_feat_global: int = config.pop("dim_feat_global", 768)
self.dim_feat_local: int = config.pop("dim_feat_local", 384)
if not is_train:
# Disable dataset caching
logger = logging.getLogger(utils.LOGGER_NAME)
logger.debug("Disable dataset caching during validation.")
config["dataset_val"]["preload_vid_feat"] = False
config["dataset_val"]["preload_text_feat"] = False
try:
self.train = RetrievalTrainConfig(config.pop(Conf.TRAIN))
self.val = RetrievalValConfig(config.pop(Conf.VAL))
self.dataset_train = RetrievalDatasetConfig(
config.pop(Conf.DATASET_TRAIN))
self.dataset_val = RetrievalDatasetConfig(
config.pop(Conf.DATASET_VAL))
self.logging = trainer_configs.BaseLoggingConfig(
config.pop(Conf.LOGGING))
self.saving = trainer_configs.BaseSavingConfig(
config.pop(Conf.SAVING))
self.optimizer = optimization.OptimizerConfig(
config.pop(Conf.OPTIMIZER))
self.lr_scheduler = lr_scheduler.SchedulerConfig(
config.pop(Conf.LR_SCHEDULER))
self.model_cfgs = {}
for key in RetrievalNetworksConst.values():
self.model_cfgs[key] = models.TransformerConfig(
config.pop(key))
except KeyError as e:
print()
print(traceback.format_exc())
print(
f"ERROR: {e} not defined in config {self.__class__.__name__}\n"
)
raise e
self.post_init()